CSCI 4717/5717 Computer Architecture

Cache Memory – Page 1 of 56CSCI 4717 – Computer Architecture

CSCI 4717/5717 Computer Architecture

Topic: Cache Memory Reading: Stallings, Chapter 4


Cache

• What is a cache? A cache is a small amount of very fast associative memory

• What makes small fast? – Simpler decoding logic– More expensive SRAM technology– Close proximity to processor – Cache sits

between normal main memory and CPU or it may be located in CPU chip or module


Cache (continued)


Cache operation – overview

• CPU requests contents of memory location• Check cache for this data• If present, get from cache (fast)• If not present, one of two things happens:

– read required block from main memory to cache then deliver from cache to CPU (cache physically between CPU and bus)

– read required block from main memory to cache and simultaneously deliver to CPU (CPU and cache both receive data from the same data bus buffer)


Going Deeper with Principle of Locality

• Cache "misses" are unavoidable, i.e., every piece of data and code must be missed so it can be loaded at least once

• What does a processor do during a miss? It waits for the data to be loaded.

• Power consumption varies linearly with clock speed and the square of the voltage.

• Adjusting clock speed and voltage of processor has the potential to produce cubic (cubed root) power reductions (http://www.visc.vt.edu/~mhsiao/papers/pacs00ch.pdf)


Cache Structure

• Cache includes tags to "associate" the source address of the block from main memory to the block contained in a line of the cache

• Each word in main memory has a unique n-bit address

• There are M=2n/K blocks of K words in main memory

• Cache contains C lines of K words each with a tag uniquely identifying where the block of K words came from


Cache Structure (continued)Line

number012

C-1

Tag Block

Block length(K words)


Memory Divided into Blocks

Block of K words

Block

Word length

Memory Address

123

2n-1


Cache Design

• Size• Mapping Function• Replacement Algorithm• Write Policy• Block Size• Number of Caches


Cache size

• Cost – More cache is expensive• Speed

– More cache is faster (up to a point) – Larger decoding circuits slow up a cache– Algorithm is needed for mapping main

memory addresses to lines in the cache. This takes more time than just a direct RAM


Typical Cache Organization


Mapping Functions

• A mapping function is the method used to locate a memory address within a cache

• It is used when copying a block from main memory to the cache and it is used again when trying to retrieve data from the cache

• There are three kinds of mapping functions– Direct– Associative– Set Associative


Cache Example

These notes use an example of a cache to illustrate each of the mapping functions. The characteristics of the cache used are:– Size: 64 kByte– Block size: 4 bytes – i.e. the cache has 16k

(214) lines of 4 bytes– Address bus: 24-bit– i.e., 16M bytes main

memory divided into 16M/4 = 4M blocks of 4 words each


Direct Mapping Traits• Each block of main memory maps to only one cache

line – i.e. if a block is in cache, it will always be found in the same place

• Line number is calculated using the following function

i = j modulo m

where

i = cache line number

j = main memory block (not full address) number

m = number of lines in the cache


Direct Mapping Address Structure

Each main memory address can by divided into three fields• Least Significant w bits identify unique word within a block• Remaining bits (s) specify which block in memory. These are

divided into two fields– Least significant r bits of these s bits identifies which line in

the cache– Most significant s-r bits uniquely identifies the block within a

line of the cache

s-r bits r bits w bits

Tag Bits identifyingrow in cache

Bits identifying wordoffset into block


Direct Mapping Address Structure(continued)

• Why are the r-bits used to identify which line in cache?

• More likely to have unique r bits than s-r bits based on principle of locality of reference


Direct Mapping Address Structure Example

• 24 bit address• 2 bit word identifier (4 byte block)• 22 bit block identifier• 8 bit tag (=22–14)• 14 bit slot or line• No two blocks in the same line have the same tag• Check contents of cache by finding line and

comparing tag

Tag s-r

8 14 2

Line or slot r Word w


Direct Mapping Cache Line Table

Cache line Main Memory blocks held

0 0, m, 2m, 3m…2s–m

1 1, m+1, 2m+1…2s–m+1

m–1 m–1, 2m–1, 3m–1…2s–1


Direct Mapping Cache Organization


Direct Mapping Examples

What cache line number will the following addresses be stored to, and what will the minimum address and the maximum address of each block they are in be if we have a cache with 4K lines of 16 words to a block in a 256 Meg memory space (28-bit address)?

a.) 9ABCDEF16

b.) 123456716

Tag s-r

12 12 4

Line or slot r Word w


More Direct Mapping ExamplesAssume that a portion of the tags in the cache in our example looks like the table below. Which of the following addresses are contained in the cache?

a.) 438EE816 b.) F18EFF16 c.) 6B8EF316 d.) AD8EF316

Addresses wi/ block Tag (binary) Line number (binary) 00 01 10 11

0101 0011 1000 1110 1110 10 1110 1101 1000 1110 1110 11 1010 1101 1000 1110 1111 00 0110 1011 1000 1110 1111 01 1011 0101 1000 1110 1111 10 1111 0001 1000 1110 1111 11


Direct Mapping Summary

• Address length = (s + w) bits• Number of addressable units = 2s+w words or

bytes• Block size = line width = 2w words or bytes• Number of blocks in main memory = 2s+ w/2w = 2s

• Number of lines in cache = m = 2r

• Size of tag = (s – r) bits


Direct Mapping pros & cons

• Simple• Inexpensive• Fixed location for given block –

If a program accesses 2 blocks that map to the same line repeatedly, cache misses are very high (thrashing)


Associative Mapping Traits

• A main memory block can load into any line of cache

• Memory address is interpreted as:– Least significant w bits = word position within block– Most significant s bits = tag used to identify which

block is stored in a particular line of cache

• Every line's tag must be examined for a match• Cache searching gets expensive and slower


Associative Mapping Address Structure Example

• 22 bit tag stored with each 32 bit block of data• Compare tag field with tag entry in cache to

check for hit• Least significant 2 bits of address identify which

of the four 8 bit words is required from 32 bit data block

Tag – s bits (22 in example)

Word – w bits(2 in ex.)


Fully Associative Cache Organization


Fully Associative Mapping Example

Assume that a portion of the tags in the cache in our example looks like the table below. Which of the following addresses are contained in the cache?

a.) 438EE816 b.) F18EFF16 c.) 6B8EF316 d.) AD8EF316

Addresses wi/ block Tag (binary) 00 01 10 11

0101 0011 1000 1110 1110 10 1110 1101 1100 1001 1011 01 1010 1101 1000 1110 1111 00 0110 1011 1000 1110 1111 11 1011 0101 0101 1001 0010 00 1111 0001 1000 1110 1111 11


Associative Mapping Summary

• Address length = (s + w) bits• Number of addressable units = 2s+w words or

bytes• Block size = line size = 2w words or bytes• Number of blocks in main memory = 2s+ w/2w = 2s

• Number of lines in cache = undetermined• Size of tag = s bits


Set Associative Mapping Traits

• Address length is s + w bits• Cache is divided into a number of sets, v = 2d

• k blocks/lines can be contained within each set• k lines in a cache is called a k-way set

associative mapping• Number of lines in a cache = v•k = k•2d

• Size of tag = (s-d) bits


Set Associative Mapping Traits (continued)

• Hybrid of Direct and Associativek = 1, this is basically direct mappingv = 1, this is associative mapping

• Each set contains a number of lines, basically the number of lines divided by the number of sets

• A given block maps to any line within its specified set – e.g. Block B can be in any line of set i.

• 2 lines per set is the most common organization. – Called 2 way associative mapping– A given block can be in one of 2 lines in only one specific

set– Significant improvement over direct mapping


K-Way Set Associative Cache Organization


How does this affect our example?

• Let’s go to two-way set associative mapping• Divides the 16K lines into 8K sets• This requires a 13 bit set number• With 2 word bits, this leaves 9 bits for the tag

• Blocks beginning with the addresses 00000016, 00800016, 01000016, 01800016, 02000016, 02800016, etc. map to the same set, Set 0.

• Blocks beginning with the addresses 00000416, 00800416, 01000416, 01800416, 02000416, 02800416, etc. map to the same set, Set 1.


Set Associative Mapping Address Structure

• Note that there is one more bit in the tag than for this same example using direct mapping.

• Therefore, it is 2-way set associative

• Use set field to determine cache set to look in

• Compare tag field to see if we have a hit

Tag 9 bits

Set 13 bits

Word 2 bits


Set Associative Mapping Example

• What cache set number will the block be stored to?• What will their tag be?• What will the minimum address and the maximum address of

each block they are in be?

1.9ABCDEF16

2.123456716

For each of the following addresses, answer the following questions based on a 2-way set associative cache with 4K lines, each line containing 16 words, with the main memory of size 256 Meg memory space (28-bit address):

Tag s-r

13 11 4

Set s Word w


Set Associative Mapping Summary

• Address length = (s + w) bits• Number of addressable units = 2s+w words or bytes• Block size = line size = 2w words or bytes• Number of blocks in main memory = 2s+ w/2w = 2s

• Number of lines in set = k• Number of sets = v = 2d

• Number of lines in cache = kv = k * 2d

• Size of tag = (s – d) bits


Replacement Algorithms

• There must be a method for selecting which line in the cache is going to be replaced when there’s no room for a new line

• Hardware implemented algorithm (speed)• Direct mapping

– There is no need for a replacement algorithm with direct mapping

– Each block only maps to one line– Replace that line


Associative & Set Associative Replacement Algorithms

• Least Recently used (LRU)– Replace the block that hasn't been touched in the

longest period of time– Two way set associative simply uses a USE bit.

When one block is referenced, its USE bit is set while its partner in the set is cleared

• First in first out (FIFO) – replace block that has been in cache longest


Associative & Set Associative Replacement Algorithms (continued)

• Least frequently used (LFU) – replace block which has had fewest hits

• Random – only slightly lower performance than use-based algorithms LRU, FIFO, and LFU


Writing to Cache

• Must not overwrite a cache block unless main memory is up to date

• Two main problems:– If cache is written to, main memory is invalid or if

main memory is written to, cache is invalid – Can occur if I/O can address main memory directly

– Multiple CPUs may have individual caches; once one cache is written to, all caches are invalid


Write through

• All writes go to main memory as well as cache

• Multiple CPUs can monitor main memory traffic to keep local (to CPU) cache up to date

• Lots of traffic

• Slows down writes


Write back

• Updates initially made in cache only• Update bit for cache slot is set when update

occurs• If block is to be replaced, write to main memory

only if update bit is set• Other caches get out of sync• I/O must access main memory through cache• Research shows that 15% of memory references

are writes


Multiple Processors/Multiple Caches

• Even if a write through policy is used, other processors may have invalid data in their caches

• In other words, if a processor updates its cache and updates main memory, a second processor may have been using the same data in its own cache which is now invalid.


Solutions to Prevent Problems with Multiprocessor/cache systems

• Bus watching with write through – each cache watches the bus to see if data they contain is being written to the main memory by another processor. All processors must be using the write through policy

• Hardware transparency – a "big brother" watches all caches, and upon seeing an update to any processor's cache, it updates main memory AND all of the caches

• Noncacheable memory – Any shared memory (identified with a chip select) may not be cached.


Line Size

• There is a relationship between line size (i.e., the number of words in a line in the cache) and hit ratios

• As the line size (block size) goes up, the hit ratio could go up due to more words available to the principle of locality of reference

• As block size increases, however, the number of blocks goes down, and the hit ratio will begin to go back down after a while

• Lastly, as the block size increases, the chances of a hit to a word farther from the initially referenced word goes down


Multi-Level Caches

• Increases in transistor densities have allowed for caches to be placed inside processor chip

• Internal caches have very short wires (within the chip itself) and are therefore quite fast, even faster then any zero wait-state memory accesses outside of the chip

• This means that a super fast internal cache (level 1) can be inside of the chip while an external cache (level 2) can provide access faster then to main memory


Unified versus Split Caches

• Split into two caches – one for instructions, one for data• Disadvantages

– Questionable as unified cache balances data and instructions merely with hit rate.

– Hardware is simpler with unified cache• Advantage

– What a split cache is really doing is providing one cache for the instruction decoder and one for the execution unit.

– This supports pipelined architectures and other mechanisms intended to improve performance.

– Instruction cache doesn't need to worry about writes.


Intel x86 caches

• 80386 – no on chip cache• 80486 – 8k using 16 byte lines and four-way set

associative organization (main memory had 32 address lines – 4 Gig)

• Pentium (all versions)– Two on chip L1 caches– Data & instructions


Pentium 4 L1 and L2 Caches

• L1 cache– 8k bytes– 64 byte lines– Four way set associative

• L2 cache – Feeding both L1 caches– 256k– 128 byte lines– 8 way set associative


Pentium 4 (Figure 4.13)


Pentium 4 Operation – Core Processor

• Fetch/Decode Unit– Fetches instructions from L2 cache– Decode into micro-ops– Store micro-ops in L1 cache

• Out of order execution logic– Schedules micro-ops– Based on data dependence and resources– May speculatively execute


Pentium 4 Operation – Core Processor (continued)

• Execution units– Execute micro-ops– Data from L1 cache– Results in registers

• Memory subsystem – L2 cache and systems bus


Pentium 4 Design Reasoning

• Decodes instructions into RISC like micro-ops before L1 cache

• Micro-ops fixed length – Superscalar pipelining and scheduling

• Pentium instructions long & complex• Performance improved by separating decoding from

scheduling & pipelining – (More later – ch14)


Pentium 4 Design Reasoning (continued)

• Data cache is write back – Can be configured to write through

• L1 cache controlled by 2 bits in register– CD = cache disable– NW = not write through– 2 instructions to invalidate (flush) cache and write

back then invalidate


Power PC Cache Organization

• 601 – single 32kb 8 way set associative• 603 – 16kb (2 x 8kb) two way set associative• 604 – 32kb• 610 – 64kb• G3 & G4

– 64kb L1 cache – 8 way set associative– 256k, 512k or 1M L2 cache – two way set associative


PowerPC G4 (Figure 4.14)


Comparison of Cache Sizes (Table 4.3)

Documents

CSCI 4717/5717 Computer Architecture